Descript Convert to AI Speech: Re-Voice a Recording (2026)
Vorec Team · 2026-10-15 · About 7 min read
The take was fine. You clicked the right things in the right order. The audio is the problem: a fan in the background, a cough, a sentence you flubbed, and you can't get the same demo environment back to record it again.
Descript's changelog now names that exact situation. In its 17 September 2026 release roundup, Descript added a way to give a recording a different voice: replace the audio on a recording or screen share with an AI voice while the script stays the same. Its next roundup, titled 12 October 2026 and posted on 13 October, then moved Descript's AI Speech to ElevenLabs v4.
This post covers what Descript's own changelog says, the steps it gives, and then our reading of what the trend means for anyone making screen-recorded tutorials. Where a point is our interpretation rather than Descript's statement, we say so.
Checked on 15 October 2026. The source is Descript's official changelog, specifically the release roundup of 17 September 2026 and the roundup titled 12 October 2026 (shown as posted on 13 October). Descript updates its product often, so treat the changelog and its help centre as the source of truth.
What changed, in Descript's words
| Roundup | Change | What the changelog says |
|---|---|---|
| 17 Sep 2026 | Give a recording a different voice | "Replace the audio on a recording with an AI voice. Your script stays the same; Descript reads it back in the voice you pick." |
| 17 Sep 2026 | A real voice library | Anywhere Descript asks you to pick a voice, the list "now comes from the ElevenLabs voice library instead of our own short set" |
| 17 Sep 2026 | Generate music and sound effects | Describe the audio you need and Descript's AI places it on the timeline at the length you asked for |
| 12 Oct 2026 | AI Speech on ElevenLabs v4 | "Voice clones sound closer to the person they were cloned from." Pick v4 in AI model settings; some projects switch on their own the next time you open them |
| 12 Oct 2026 | Recording requests | Send a link so someone records on their own schedule, optionally reading a script from a teleprompter. Descript lists this as available on Enterprise drives |
The changelog frames the re-voicing feature around a specific case: it is "useful when the take is fine, the audio isn't, and you can't record it again."
How to re-voice a recording in Descript
These are the steps from Descript's changelog, condensed:
- Select the recording or screen share you want to replace.
- Choose Convert to AI Speech.
- Pick a voice.
That's the whole flow as the changelog describes it. Two practical notes follow from the description, and both are our reading rather than Descript's instructions:
- The script is the transcript you already have. Because Descript reads back "your script", fix any wording in the transcript before converting. The AI voice says what the script says, mistakes included.
- Choose the voice deliberately. Since the 17 September update, the picker draws on the ElevenLabs voice library. A bigger library makes it easier to match a brand voice and easier to pick one at random. If your team publishes a series, settle on one voice and write it down.
This feature entry does not say which plans include Convert to AI Speech or how many AI credits it uses, so check that against your own plan. Our Descript pricing breakdown covers the plan structure as we last verified it.
Why this matters for tutorial makers
This section is our analysis, not Descript's.
Recording a screen and recording a voice are two different skills, and for years tutorial tools treated them as one take. You clicked and talked at the same time, and any slip in either meant starting over. The features in these two release roundups pull the two apart: keep the visual take, regenerate the voice.
That split is the trend worth watching, and it is not unique to one product. We cover the voice side broadly in our AI voiceover guide and the cloning side in AI voice cloning explained. What Descript's update makes concrete is the use case. The changelog names screen shares specifically, which is the format most software tutorials are recorded in.
This announcement describes re-voicing a script you already have. It does not describe generating narration from the on-screen actions of a silent recording, so we don't assume either way. Here is how the common starting points compare in general, whatever tool you use:
| Starting point | What you still have to do |
|---|---|
| You narrated live and the audio is bad | Fix the transcript, then re-voice it. This is the case Descript's feature targets |
| You narrated live and the words are wrong | Rewrite the script, then re-voice it. The visuals may no longer match the new words |
| You recorded silently, with no narration at all | Get a script that matches what happens on screen, written by hand or with a tool that drafts one, then voice it |
| The process changed and the screen looks different | Re-record the visuals. A new voice cannot fix outdated screens |
The third row is the one silent screen recorders know well. If you never spoke, the re-voicing step described in this announcement needs a script to read, and that script has to line up with each click on screen. Our guides to making software tutorials without a mic and what a no-mic tutorial is cover that workflow.
A checklist before you re-voice a tutorial
Whatever tool you use, these steps keep an AI voiceover honest to the recording underneath it:
- Watch the take once with the sound off. If the visuals have a mistake, a new voice won't fix it.
- Edit the script for the viewer, not the recorder. Live narration is full of "so, um, now I'm going to". A regenerated voice is a chance to say "Open Settings" instead.
- Check timing after conversion. An AI voice reads at its own pace, so confirm each sentence still lands while the matching action is on screen.
- Keep one voice per series, and record which one it is.
- Use a cloned voice only with permission. Descript's update improves clone quality; the consent question is yours to settle before you clone anyone, including yourself on a shared account.
Where Vorec fits
Vorec starts from that third row. It records your screen, with its own macOS recorder or from a recording you upload, and drafts narration matched to the workflow it captured, then generates the voiceover, so nothing is spoken into a microphone. In the editor you can add smooth cursor motion, cursor-follow zoom and click-based auto-zoom, and freeze-sync can hold a frame to give an explanation time to finish. The same capture can also produce a written step-by-step guide with annotatable screenshots, and narration regenerates in supported languages on eligible plans without re-recording.
An AI agent can drive the recording too. Vorec's `record-tutorial` skill lets Claude Code or Codex plan a walkthrough and drive the Vorec recorder through the CLI; the capture stays local for you to review before anything is uploaded and narrated. Our post on the Claude Code screen recording plugin covers that workflow. If you already work in Descript and mostly need to rescue spoken takes, its re-voicing feature is aimed squarely at that. If you mostly record silently, Vorec is built around writing the narration for you.
Plans are listed on the Vorec pricing page. Trial includes up to 3 projects; exports carry a watermark.
FAQ
What is Convert to AI Speech in Descript?
It is the option Descript's 17 September 2026 changelog describes as giving a recording a different voice. You select a recording or screen share, choose Convert to AI Speech and pick a voice, and Descript reads your existing script back in that voice.
Does it keep my original words?
Yes, according to the changelog: "Your script stays the same." Edit the transcript first if you want the AI voice to say something different.
What changed with ElevenLabs v4?
Descript's roundup titled 12 October 2026 says AI Speech now runs on ElevenLabs v4 and that voice clones sound closer to the person they were cloned from. You pick v4 in AI model settings, and some projects switch automatically when reopened.
Can I use it on a silent screen recording?
The announcement describes reading back an existing script in a new voice. On its own, it doesn't establish whether Descript can draft narration from a silent recording's on-screen actions; check Descript's help centre for that workflow.
Which Descript plans include it?
The changelog entry for this feature does not say. Check Descript's pricing page or help centre for your plan.
Is Descript's recording request feature on every plan?
No. The roundup titled 12 October 2026 lists recording requests as available on Enterprise drives.
Want the recording to narrate itself? Record with Vorec, or upload one you already have, and get a narrated tutorial plus a written guide. Start free. 7-day trial, 100 credits, no credit card required. Trial includes up to 3 projects; exports carry a watermark.